Forum

Author Topic: depth filtering processing time  (Read 5422 times)

ferdywallinx

  • Newbie
  • *
  • Posts: 28
    • View Profile
depth filtering processing time
« on: January 19, 2016, 02:08:10 PM »
Hi All,

I am currently running quite big datasets on my computer and noticed that quite often there is a big time lag in processing time during the depth filtering step. In the attachment you can see a plot of time related to camera count and I would like to know three things:

1: can I expect that with, say,  600 camera's my system will never finish?
2: what parameters can I set to control the number of camera's used in a substep in the depth filtering step?
3: how can i further decrease this processing time?

I'd like to note that my system is not swapping and the RAM is only 30% used. the 32 CPU cores are running full-time (XEON E-5 Dual @ 3.5GHz)

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15282
    • View Profile
Re: depth filtering processing time
« Reply #1 on: January 19, 2016, 02:20:28 PM »
Hello ferdywallinx,

Depth filtering may take a long time due to excessive overlap. Usually the process is slowed significantly when there are overview photos - images that overlap with tens and even hundreds of other images. For example, for close range objects it means that there are mostly close-up images but some photos include the complete object in the frame. Or if speaking of aerial survey, the images are taken from several heights and the ones taken from the higher altitude have lower GSD, resulting in the overlap with considerable amount of lower altitude photos.
Best regards,
Alexey Pasumansky,
Agisoft LLC

Arie

  • Full Member
  • ***
  • Posts: 134
    • View Profile
Re: depth filtering processing time
« Reply #2 on: January 19, 2016, 02:24:02 PM »
Hi,
1. it depends on the resolution of the images and the quality setting used for reconstruction. Furthermore the reconstruction time also depends on the type of scene.
"Never" is quite a long time..

2. Afaik, none.

3. Use a lower quality setting for reconstruction or divide your project into seperate chunks and merge them afterwards. Check the manual for how to use chunks. Other than that the only thing i can thin of is more powerful hardware or node setup.

Cheers!

Edit: Alexey, thats interesting! Thanks for the info. So in regards to shooting strategy would it be wise to capture some overview images for higher accuracy in alignment and deactivating them for depth reconstruction?
« Last Edit: January 19, 2016, 02:27:13 PM by Arie »

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 15282
    • View Profile
Re: depth filtering processing time
« Reply #3 on: January 19, 2016, 03:06:11 PM »
Hello Arie,

If there's no overlap issues, I can suggest to exclude those general overview images from the dense cloud generation process. Usually they are shot in lower effective resolution, so actually do not provide any quality improvement.
Best regards,
Alexey Pasumansky,
Agisoft LLC